ABSTRACT
Knowledge acquisition, knowledge representation, and inference are common terms in the artificial intelligence (AI) literature; they are the central elements of a knowledge-based (or expert) system. AI, however, is not alone in its concern with these issues. Decision analysis (DA) and other statistical decision sciences have also investigated the elicitation of information from human experts and the formal modelling of this information to infer useful recommendations. The approach traditionally taken by AI has been quite different from that favoured by DA. Perhaps the most fundamental distinction between them has been their psychological motivation. Many of the procedures popular in AI are rooted in descriptive psychology, or observations about the way that people behave. The prevalent paradigm for expert system design is a case in point; production rules attempt to model the problem-solving strategies employed by experts. In other words, an ideal rule-based system would mimic the behaviour of a human expert. DA, on the other hand, is a sophisticated outgrowth of task analyses of inferences, evaluations and decisions. In so far as it relates to human behaviour, DA prescribes what decision makers should do, rather than describing what they do. Decision-analytic task analyses often lead to prescriptions that differ systematically from what people do (see, for example, Kahneman el al., 1982); such effects have been called cognitive illusions, as in von Winterfeldt and Edwards (1986).
